Jieyuan Pei

World4Scorer architecture: a latent world model predicts a state for each candidate trajectory and score heads rate each one; training pairs a future loss on the logged plan with simulator outcome losses

arXiv:2609.36438

World4Scorer: Outcome-Grounded World Modeling for Autonomous Driving

Jieyuan Pei^*, Meiyi Lu^*, Sining Ang, Yubo Zhao, Zhangyi Hu, Mingwei Xu, Haokai Ding, Wei Li, Zihan You, Jianwei Zheng, Li Yu, Yifeng Pan, Ji Tao, Rongjunchen Zhang, Yan Wang^†

93.0 EPDMS on NAVSIM-v2, 94.0 PDMS on NAVSIM-v1 and a 73.27 driving score on Bench2Drive.

arXiv → Project page → Code →

HNO teaser: near/far-field decomposition, hierarchical tree, and its embedding on a Lorentz hyperboloid

ICML 2026

Hyperbolic Neural Operator

Jieyuan Pei, Zhuoxuan Li, Wei Li, Haobo Zhang, Jiawei Jiang, Jianwei Zheng^†

Lorentz-distance attention for multiscale near–far interactions in PDE surrogates, evaluated on six standard benchmarks and two unstructured-mesh CFD tasks.

Paper PDF → Abstract & text → Project page → Code →

S3O architecture: multi-scale embedding, S3O backbone, and the hard location selection block

CVPR 2026

S³O: Selective Spatial–Spectral Operator for Cross-Scale Fusion

Jieyuan Pei^*, Wei Li^*, Zhuoxuan Li, Junwei Zhu, Meiyi Lu, Jiawei Jiang, Chenyu Wang, Jianwei Zheng^†

0.3–1.4 dB gains on three datasets at up to 4× lower compute than heavy Transformer / Mamba baselines.

Paper → Code →

StarFus-Net U-Net with asymmetric spatial measurement and spectral prior paths

ICASSP 2026

Asymmetric StarFus: Learning Incoherent Measurements for Semantics-Aware Spatial–Spectral Fusion

Jieyuan Pei, Wei Li, Zhuoxuan Li, Junwei Zhu, Chuangjie Fang, Jianwei Zheng^†

Paper → Code →

GTR-Mamba architecture: rotation, geo and time encoders feeding GTR SSM layers with manifold-tangent recurrence on the Lorentz model

arXiv:2510.22942

GTR-Mamba: Geometry-to-Tangent Routing Mamba for Hyperbolic POI Recommendation

Zhuoxuan Li^*, Jieyuan Pei^*, Tangwei Ye, Zhongyuan Lai, Zihan Liu, Fengyuan Xu, Qi Zhang, Liang Hu^†

arXiv →

Latent spectral operator pipeline: patchify, coordinate-to-latent projection, solve on a trigonometric basis, project back

ICML 2026

Solving Spatial–Spectral Fusion with Latent Spectral Operators

Wei Li, Jieyuan Pei, Junnan Xu, Xuanfeng Ding, Junwei Zhu, Wanjun Chen, Jianwei Zheng^†

+0.160 PSNR on CAVE ×4/×8 with 4.4% / 6.9% SAM reduction.

Paper → Project page → Code →

Five-stage fusion pipeline with parallel spatial and frequency branches and a semantic transformer

IEEE TGRS

S²TA-Fuse: Semantic-Superpixel Tokenized Attention for Spatial–Spectral Fusion

Jiawei Jiang^*, Wei Li^*, Jieyuan Pei, Junwei Zhu, Honghui Xu, Yuchao Feng^†, Jianwei Zheng^†

Paper →

Mag-Mamba framework: magnetic phase encoder feeding a Mamba layer with rotation control

KDD 2026 · Oral

Mag-Mamba: Modeling Coupled Spatiotemporal Asymmetry for POI Recommendation

Zhuoxuan Li^*, Tangwei Ye^*, Jieyuan Pei, Haina Liang, Zhongyuan Lai, Zihan Liu, Yiming Wu, Qi Zhang, Liang Hu^†

arXiv →

RoboMemArena overview: annotation pipeline, dataset statistics, baseline comparison, and real-world evaluation

arXiv:2605.10921

RoboMemArena: A Comprehensive and Challenging Robotic Memory Benchmark

Huashuo Lei^*, Wenxuan Song^*, Huarui Zhang, Jieyuan Pei, Jiayi Chen, Haodong Yan, Han Zhao, Pengxiang Ding, Zhipeng Zhang, Lida Huang, Donglin Wang, Yan Wang, Haoang Li^†

26 tasks, 2,600 demonstrations, 2.8M frames; mean trajectory 1,076 steps.

arXiv → Project page → Code →

Researchthree threads

AI for Science

Autonomous Driving

Multimodal Models

Papersbold = me · equal contribution · † corresponding author · click a panel for the full figure*

Experiencewhere the work happens

Awardsnational level